Digital communications are widely used for the transmission of voice, data and video information. Such transmission can extend over large geographical distances, between components within a personal computer, or only between adjacent circuit portions on an integrated circuit. Certain such communications applications benefit from or require the conversion of serial data into parallel data for simultaneous transmission over parallel communications channels, or more generically, from M'ary symbols to N'ary symbols. At the receiving end, the parallel data is desirably converted back into the serial data, and with the bits or symbols in the correct order to avoid data errors.
Unfortunately, the demand for greater data transmission volumes and at ever higher speeds, may result in skew at the receiver. In other words, the parallel communications channels may introduce different delays to the parallel symbol strings they carry. Because of skew, the parallel symbol strings at the receiver can then no longer be simply reassembled into the starting data.
The skew problem with parallel communications channels has been addressed in a number of ways. For example, U.S. Pat. No. 4,677,618 to Haas et al. recognized the dispersion introduced by wavelength division multiplexed communications channels over optical fiber. This patent discloses determining the relative delays between the channels based upon detecting two bits in a given byte of data. The relative times of arrival of the remaining bits in a byte are predetermined using the relative delay between the two detected bits and the known frequency-related dispersion characteristics of the transmission medium. Certain bits in each received byte may then be delayed using clock delay lines or registers, thereby accounting for skew.
Along similar lines, U.S. Pat. No. 5,157,530 to Loeb et al. also determines and accounts for skew imparted by dispersion in fiber optic wavelength division multiplexing. Relative delays are used to control adjustable delay devices in each channel.
U.S. Pat. No. 5,408,473 to Hutchinson et al. is directed to a technique for synchronizing run-length-limited data transmitted over parallel communications channels. Block boundary synchronization is established during connection initialization by using a property of a required HALT code to detect block boundaries received in each channel. Skew compensation is effected by comparing the times of detection of the block boundaries in the two channels, and appropriately controlling a variable delay in at least one of the channels. If there is a subsequent loss of synchronization, detected transmission errors will eventually result in connection reinitialization and reestablishment of synchronization. Unfortunately, the transmission of the fixed HALT code to detect boundaries may result in false boundary detection. Moreover, since synchronization is not continuously maintained, the technique may be impractical for higher data rates.
U.S. Pat. No. 5,793,770 to St. John et al. is directed to a high-performance parallel interface (HIPPI) to a synchronous optical network (SONET) gateway, and wherein electronic logic circuitry formats data and overhead signals into a data frame for transmission over a fiber optic channel. Stripe skew adjustment is based upon SONET framing, and, as such, the circuitry is relatively complicated, comprising as many as 20,000 logic gates, for example.
The difficulty with skew caused by parallel communications channels is also an important issue to be addressed in communications channels between integrated circuit devices. For example, higher transmission speeds increase the sensitivity to skew, as there is a smaller time window to correctly identify a received bit and have it properly align with bits received on the other parallel communications channels. To provide a higher aggregate transmission rate, the number of parallel communications channels can be increased, without increasing the speed of any given communications channel. However, this may result in significant costs for the additional communications channels. Moreover, for communications between integrated circuits, increasing the number of communications channels increases the number of pins needed for connecting the IC. The number of pins and additional packaging complexity may significantly increase the costs of such approaches.
For communications channels between physical layer devices (PLDs) or PHY devices, and logical link devices (LLDs), typical interfaces are asymmetrical and the devices are operated in a push-pull configuration. Because of the asymmetry, relatively expensive memory is required on the PLD since it is polled by the LLD, such as an asynchronous transfer mode (ATM) device. Further developments and improvements in the communications interface between a PLD and LLD are also hampered by the skew difficulty described above as a result of higher bit rates over limited parallel communications channels.